Pantanal Land Use Land Cover Mapping with LSTM and Landsat-8/OLI Time Series

Jupyter notebooks and final map for the Land Use Land Cover mapping of Brazilian Pantanal.

Files:

• Gap Filling CubicSpline.ipynb
  • Jupyter notebook containing the scripts used to fill the gaps caused by cloud and cloud shadows in Landsat-8/OLI imagery. The method interpolates the missing values with a CubicSpline temporal function.
• Train LSTM tenfold.ipynb
  • Jupyter notebook containing a tenfold cross validation rpocess that trains 5 different models using raw or filled time series samples.
• Predict suing LSTM trained model.ipynb
  • Jupyter notebook containing the script used to employ a trained model to predict over data for the whole Pantanal.
• Pantanal_LULC.tif
  • The final pantanal LULC classfication. Class codes:
    • 1 - Water
    • 2 - Forest
    • 3 - Non Forest Natural Formation
    • 4 - Farming
    • 5 - Non Vegetated Areas